The Rewards to Meeting or Beating Earnings Expectations

Eli Bartov, Dan Givoly and Carla Hayn*

March 1999 Current Version: October 2000 We thank Michael Brennan, Jack Hughes, Patricia Hughes, Jim Ohlson and Joshua Ronen for their helpful comments on this paper. The capable computer assistance provided by Ashok Natarajan is appreciated. I/B/E/S International, Inc. provided data on analysts’ earnings forecasts. * The authors are, respectively, at New York University, University of California at Irvine, and University of California at Los Angeles.

The Rewards to Meeting or Beating Earnings Expectations

Abstract

The paper studies the manner by which earnings expectations are met, measures the rewards to

meeting or beating earnings expectations (MBE) formed just prior to the release of quarterly

earnings, and tests alternative explanations for this reward. The evidence supports the claims that the

MBE phenomenon has become more widespread in recent years and that the pattern by which MBE

is obtained is consistent with both earnings management and expectation management. More

importantly, the evidence shows that after controlling for the overall earnings performance in the

quarter, firms that manage to meet or beat their earnings expectations enjoy an average quarterly

return that is higher by almost 3% than their peers that fail to do so. While investors appear to

discount MBE cases that are likely to result from expectation or earnings management, the premium

in these cases is still significant. Finally, the results are consistent with an economic explanation for

the premium placed on earnings surprises, namely that MBE are informative of the firm’s future

performance.

Key Words: Earnings expectations, analysts’ forecasts, expectation management, earnings management, losses

The Rewards to Meeting or Beating Earnings Expectations

1. Introduction

Meeting or beating analysts’ forecasts of earnings is a notion well entrenched in today’s

corporate culture. From corporate boards’ deliberations to financial press reporting and Internet chats,

emphasis is placed on whether the company meets its earnings forecasts. The following comment

typifies the view of the financial press regarding the importance of meeting Wall Street’s expectations:

“In January, for the 41st time in 42 quarters since it went public, Microsoft reported earnings that meet or beat Wall Street estimates…. This is what chief executives and chief financial officers dream of: quarter after blessed quarter of not disappointing Wall Street. Sure, they dream about other things… But the simplest, most visible, most merciless measure of corporate success in the 1990s has become this one: Did you make your earnings last quarter?” (See Fox [1997], p. 77.)

The importance assigned to meeting earnings expectations is not surprising given the valuation

relevance of earnings information. Recent anecdotal evidence, however, suggests that companies are not

merely passive observers in the game of meeting or beating contemporaneous analysts’ expectations

(hereafter referred to as MBE). Rather, they are active players who try to win the game by altering

reported earnings or managing analysts’ expectations (see for example McGee [1997] and Vickers

[1999]). The motivations often suggested for such behavior are to maximize the share price, to boost

management’s credibility for being able to meet the expectations of the company’s constituents (e.g.,

stockholders and creditors), and to avoid litigation costs that could potentially be triggered by

unfavorable earnings surprises.

The purpose of this paper is threefold. The first objective of the paper is to test whether, after

controlling for the earnings forecast error for the period, there is a market premium to firms that meet or

beat their earnings expectations formed just prior to the release of quarterly earnings. The second

objective is to determine whether the data on earnings forecast revisions and earnings surprises are

consistent with expectation or earnings management and, if so, whether the market premium to MBE

exists even in cases where earnings or expectations are likely to have been managed. Finally, the paper

attempts to provide explanations for the potential payoffs from an MBE strategy that are consistent with

investor rationality.

Based on a sample of over 150,000 quarterly earnings forecasts made between the years 1983 to

1997 and covering approximately 75,000 firm-quarters, the paper finds that, in line with previous

research, instances in which companies meet or beat contemporaneous analysts’ estimates have

2

increased considerably in recent years. The trend is common to all four quarterly reporting periods and

is present also in the annual period. It is observed for both large and small firms. On average, analysts’

forecasts made at the beginning of the period overestimate earnings (see similar findings by Barefield

and Comiskey [1975] and Brown [1997], among others). However as the end of the reporting period

approaches analysts’ optimism (i.e., their overestimation) turns, through the predominance of downward

revisions in earnings estimates, into pessimism (i.e., underestimation). Furthermore, the proportion of

negative forecast error cases (measured relative to analysts’ earnings forecasts made at the beginning of

the quarter) that ends with a zero or positive earnings surprise (measured relative to the most recent

analysts’ earnings estimate) is greater than the proportion of positive or zero forecast error cases that

ends with a negative surprise. These findings are consistent with expectation management that takes

place late in the reporting period.

Our primary findings show that investors reward firms whose earnings meet or beat analysts’

estimates. After controlling for the quarterly forecast error (measured relative to analysts’ earnings

forecasts made at the beginning of the quarter), the quarter’s abnormal returns (measured over the same

period as the corresponding quarterly forecast errors) are positively and significantly associated with the

earnings surprise for the quarter (measured relative to the most recent estimate). The average return

over quarters associated with a positive earnings surprise is significantly higher, by almost 3%, than the

return over quarters that have the same overall quarterly earnings forecast error but end with a negative

earnings surprise. These results suggest that, independent of the market valuation of the earnings level,

there is a reward (penalty) to beating (failing to meet) analysts’ earnings expectation. Furthermore, our

tests, based on accrual analysis, show that the premium to MBE is only slightly reduced when earnings

or expectations appear to have been managed.

In examining alternative explanations for the premium, our tests do not support the notion that

investors overreact to earnings surprises (see, e.g., Zarowin 1989, and DeBondt and Thaler 1990). Such

overreaction, if present, should lead to subsequent market reversals of the abnormal returns generated

by the earnings surprise. Yet our tests, based on the examination of abnormal return over both a short

window (consisting of the following quarter) and longer windows (up to three years following the

earnings announcement), do not detect such a reversal. We further rule out several sources of

measurement errors in analysts’ forecasts, the proxy used for the unobservable market expectations of

earnings, as potential explanations for the premium to earnings surprises.

The premium to earnings surprises, however, appears to be justified on economic grounds:

Earnings surprises apparently possess greater information content with respect to future earnings. This

3

is evident from the positive association between earnings surprises and future firm performance. While

the reasons for this association are not investigated here, its presence suggests that investors rationally

react to earnings surprises.

The paper is organized as follows: The next section reviews the recent research on the issue of

MBE. Section 3 presents the empirical design, followed by a description of the sample and the data.

Results are provided in Section 5. The paper concludes with a short summary and suggestions for future

research.

2. Recent studies on MBE

The phenomenon of MBE has recently attracted interest among researchers. Brown [2000] finds a

disproportional number of cases in recent years where earnings per share are slightly (by a few cents)

above analysts’ forecasts. He further finds an increase over the years in the number of cases where

actual earnings per share are exactly on target. Degeorge et al. [1999] ascertain that the MBE strategy is

one of three performance thresholds that management tries to meet. Evidence provided by other studies

suggests that both earnings manipulation and expectation management are used to meet that end.

Burgstahler and Eames [1998] provide evidence that downward revisions of forecasts occur more

frequently when the revision would be sufficient to avoid a negative earnings surprise, suggesting

managers’ influence on analysts’ forecast revisions. Further, they conclude that the time-series behavior

of earnings is consistent with companies managing their earnings so as to meet analysts’ expectations.

Evidence consistent with earnings management to meet earnings forecasts is provided also by Kasznik

[1999] and Payne and Robb [1997]. On the other hand, Skinner [1995], Kasznik and Lev [1995],

Francis et al. [1994] and Soffer et al. [2000] show that companies increasingly tend to pre-warn

investors about forthcoming unfavorable earnings. That is, the MBE strategy is accomplished through

expectation management.

Whether carried out through earnings manipulation, expectation management or both, the

benefits from an MBE strategy are not immediately apparent, unless MBE acts as a predictor of the

future prospects of the firm. Specifically, for a policy of MBE carried out through earnings management

to be successful, investors must be incapable of detecting management’s reporting objectives. Likewise,

for an MBE policy achieved through expectation management to be successful, investors must be

incapable of correcting for an extractable past pattern of forecast errors.

A net reward from MBE through managing earnings expectations is questionable for yet another

reason. Dampening earnings expectations so as to preempt an expected unfavorable earnings surprise

4

would result in a negative price effect that should negate the positive announcement period return,

leaving the total return for the period unchanged. In fact, past research (see Kasznik and Lev [1995] and

Soffer et al. [2000]) shows a significant decline in the stock price of companies who pre-warn investors

about forthcoming unfavorable disclosures (thus lowering investors’ earnings expectations).

In a related study, Kasznik and McNichols [1999] examine whether MBE results in a higher

firm valuation and higher forecasted earnings. Regressing stock prices on book values, the estimated

present value of future abnormal income and a dummy variable capturing whether or not expectations

were met, they find that firms which consistently (i.e., over three successive years) meet earnings

expectations enjoy a valuation premium. Furthermore, firms that meet or beat earnings expectations

report higher earnings in future periods (after controlling for current earnings). Yet, the higher future

earnings of these firms are not fully incorporated in the earnings forecasts made by analysts for these

future periods. No significant market premium is found, however, for firms that met or exceeded

expectations only in the most recent year. Lopez and Rees [2000] find a premium to MBE: after

controlling for the magnitude of unexpected earnings, the stock return during the earnings

announcement period is affected by whether or not the analysts’ forecasts are met.

Our analysis differs from the concurrent studies by Kasznik and McNichols [1999] and Lopez

and Rees [2000] in several respects. First, by providing information on and analyzing the expectation

paths, we examine the manner by which management accomplishes the task of MBE, contributing to the

research on management of earnings expectations. In particular, our research method allows us to

distinguish between the two managerial tools for achieving MBE: earnings management and

expectations management. Second, we provide evidence on the increase in the premium to MBE over

recent years, thus suggesting an incentive for the increased frequency of the MBE over time, a

phenomenon documented by this as well as other studies. Third, we examine the relation between the

premium to MBE and the presence of expectation and earnings management. Finally, whereas Kasznik

and McNichols use a valuation test focusing on price levels, we use an information-content/event study

paradigm, focusing on returns. Concentration on the exact arrival time of information to the market (in

an event study design) regarding whether the earnings forecast has been met or exceeded improves the

power of our tests. In addition, unlike tests based on valuation models (such as that by Ohlson [1995]),

this approach avoids the measurement problems associated with estimating basic parameters such as

abnormal earnings, terminal value and the cost of capital.

Indeed, the event-study methodology provides more conclusive results. While Kasznik and

McNichols [1999] find a premium to MBE only in cases of “habitual beaters,” our findings detect such

5

a premium in all MBE cases. The excess average abnormal return in firm-quarters that end with

favorable earnings surprises relative to those in which earnings fell short of expectations is almost 3%.

3. Hypotheses Development

The following description of the chronology of the earnings forecasts and the earnings

announcements, illustrated in Figure 1, is helpful in understanding the hypotheses and tests of the paper.

Over the course of the current quarter and surrounding days, the earnings announcement of the

preceding quarter is issued and, subsequently, an earnings forecast for the current quarter is made,

forecast revisions occur and, following the end of the quarter, the actual earnings number is released.

We define the net revision in analysts’ forecasts of earnings during the quarter (REVQ), as the difference

between the latest earnings forecast available for that quarter and the earliest earnings forecast made

after the release of the preceding quarter’s earnings and (hereafter Flatest and Fearliest, respectively). We

further define the earnings surprise for the quarter (SURPQ) as the difference between the actual

earnings number for the quarter and Flatest. The forecast error for the quarter (ERRORQ) is the difference

between the actual earnings number and Fearliest.

For ease of exposition, we denote the combination of the direction of the net revision (up, down

or zero) and the sign of the earnings surprise (positive, negative or zero) as the “expectation path.”

Because the set of possible expectation paths is somewhat different for cases with positive, zero or

negative errors, we examine the paths separately for each error-sign group. Specifically, the following

paths are examined:

Error-Sign Group

Revision: Flatest - Fearliest

Surprise: EPS - Flatest

Expectations Path

Positive Forecast Errors + + Up – Up ( EPS - Fearliest > 0 ) + - Up – Down - + Down – Up + 0 Up – No Change 0 + No Change – Up Negative Forecast Errors - - Down – Down ( EPS - Fearliest < 0 ) - + Down – Up + - Up – Down - 0 Down – No Change 0 - No Change – Down Zero Forecast Errors + - Up – Down ( EPS - Fearliest = 0 ) - + Down – Up 0 0 No Change – No Change

6

Note that meeting or beating expectations corresponds to paths that end with “No Change” and “Up,”

respectively.

We conduct analyses designed to: (1) document the pattern of an increased frequency of MBE

over time, (2) test for the presence of expectation management, as evidenced by the behavior of mid-

period analysts’ revisions and (3) establish whether there is a premium associated with MBE.

3.1. Hypotheses relating to expectation management

To test for the presence of expectation management, we perform two tests. First, we contrast the

actual earnings surprise distribution with the hypothetical distribution assuming no interim revision in

analysts’ earnings forecasts. Note that the latter distribution is identical to that of the forecast error. If

expectation management occurs, specifically, if expectations are dampened leading to downward

revisions in earnings estimates, we would expect to find a lower frequency of negative earnings

surprises than negative forecast errors. Conversely, if interim forecast revisions only represent the

arrival of information without any managerial effort to manage expectations, no difference should be

observed between the frequency of negative earnings surprises and negative forecast errors.

Accordingly, the following hypothesis is tested (all hypotheses are expressed in their alternative form):

H1a: The relative frequency of negative earnings surprises is smaller than the relative frequency of negative forecast errors.

If expectation management has become more pronounced in recent years, we further expect the excess

frequency of negative forecast errors over negative earnings surprises to have increased over time.

H1b: The excess of the relative frequency of negative forecast errors over negative earnings surprises has increased over time.

In a second, related test of the expectation management hypothesis, we examine more closely

the role of the interim forecast revision in affecting the sign of the end-of-quarter earnings surprise. Our

examination is based on a comparison of the observed sign of the earnings surprise with the sign of the

earnings surprise that would have been observed in the absence of an interim forecast revision. As

explained above, in the absence of an interim revision, the sign of the earnings surprise would be the

same as the sign of the quarterly forecast error. Observing a negative forecast error that ends, through a

sufficiently large downward revision, with a positive earnings surprise is thus consistent with

expectation management. In the same vein, an observation with a zero or positive forecast error that

ends, due to a sufficiently large upward forecast revision, with a negative earnings surprise is

inconsistent with expectation management. If there is no management intervention, the proportion of

observations in which the interim forecast revision offsets the sign of the forecast error so as to change

7

the sign of the earnings surprise should be identical between cases with negative errors and cases with

positive errors. Our hypothesis is stated in terms of the difference between these proportions.

Specifically:

H2a: The proportion of negative forecast error cases that ends with a zero or positive surprise (i.e., a Down-Zero or Down-Up path) is greater than the proportion of positive or zero forecast error cases that ends with a negative surprise (i.e., an Up-Down path).

If expectation management has become more prevalent recently, we expect an increase over

time in the excess of the proportion of cases with the Down-Zero or Down-Up paths among negative

forecast errors compared with the proportion of cases with the Up-Down path among positive or zero

forecast errors. That is:

H2b: The difference between the proportion of negative forecast error cases that end with a zero or positive surprise and the proportion of positive or zero forecast error cases that end with a negative surprise has increased over time.

3.2 Hypotheses regarding the premium to MBE

If the expectation path is not informative with respect to future firm performance and investors

are rational, the course of the expectation path should not affect the abnormal return for the quarter. In

particular, there should be no reward to an MBE strategy. Accordingly, the following hypothesis is

advanced (again, expressed in its alternative form):

H3a: After controlling for the forecast error, there is a premium to MBE.

Consistent with Hypotheses 1b and 2b regarding the increasing prevalence of the MBE

phenomenon, we examine the behavior of the premium to MBE over time, by testing the following

hypothesis:

H3b: After controlling for the forecast error, the premium to MBE has increased over time.

To better understand the nature of the premium (if any), we further test two additional

hypotheses. One is that the premium to MBE and the penalty for failing to meet expectations are, per

unit of surprise, of the same magnitude. The second is that the premium to meeting expectations is

similar to that associated with beating expectations. Stated in their alternative forms, these hypotheses

are thus that, after controlling for the forecast error for the period:

H4: The premium to MBE is different from the penalty for failing to meet expectations.

H5: The premium to meeting expectations is different from the premium to beating expectations.

3.3. Hypotheses regarding factors influencing the premium to MBE

3.3.1. Earnings persistence

8

Previous research has documented a differential stock price response to earnings depending on

whether earnings are positive or negative (e.g., Hayn [1995]). Further evidence suggests that the

persistence of earnings decreases is significantly lower than the persistence of earnings increases (see

for example, Brooks and Buckmaster [1976], Elgers and Lo [1994], and Basu [1997].). If investors

perceive losses and earnings declines as transitory, then their response to MBE in these instances may

also be muted. However, if investors value MBE more in cases of losses or declining profits than in

cases of profits or increased earnings due to a reduction in expected bankruptcy costs, the premium to

MBE should be larger for the former cases. Accordingly, we test the following hypotheses (stated in

their alternative form) that, after controlling for the forecast error for the period:

H6: The premium to MBE in loss cases is different from the premium to MBE in profit cases. H7: The premium to MBE in cases of decreasing earnings is different from the premium to MBE in cases of increasing earnings.

3.3.2. MBE persistence Depending on the cause for the premium to MBE, the intensity of investors’ response to

instances of MBE may be influenced by the track record of the firm. If MBE is perceived to be a signal of future performance, repeated instances of MBE would indicate earnings momentum and produce a greater premium than isolated cases of MBE. On the other hand, an observed pattern of successive instances of MBE may indicate to investors the presence of management intervention and thus be associated with a lower premium. Persistent MBE behavior may result in a lower premium for yet another reason: It may indicate a bias in the forecasts used in this study that is unlikely to be present in an efficient market. The forecasts that we use, obtained from I/B/E/S (as described in the following section), may measure the true, unobservable, market expectations of earnings with error. This would, in turn, result in a lower or no observed premium to MBE.

The evidence provided by Kasznik and McNichols [1999] is consistent with the first prediction of a greater premium to firms that consistently meet or beat expectations. In fact, their tests fail to detect a significant premium to “one-time beaters.” The event-study methodology used here may be more powerful in detecting a market premium if it exists. The above considerations lead to the following hypothesis (expressed in alternative form):

H8: The premium to MBE of “habitual beaters” of expectations is different from the premium to MBE of “sporadic beaters.”

3.4. Hypotheses regarding the association between the premium to MBE, expectation management and earnings management MBE may be achieved in some cases through expectation management or earnings management. If investors can trace the MBE to management intervention, they may not reward such

9

cases with the same premium, or not reward them any premium at all. Accordingly, we test two additional hypotheses: H9: The premium to MBE is larger for cases that are less likely to be driven by expectation

management. H10: The premium to MBE is larger for cases that are less likely to be driven by earnings management.

4. Sample and Data

The sample consists of firm-quarter observations on the I/B/E/S database that satisfy the

following criteria:

1) There are at least two individual earnings forecasts (not necessarily by the same analyst) for the

quarter that are at least three days apart from each other.

2) The release date of the earliest forecast occurs at least three days after the release of the previous

quarter’s earnings.

3) The release date of the latest forecast precedes the earnings release by at least three days.

The first criterion ensures that there is an initial forecast and a subsequent forecast revision.

These are required to be separated in time by at least three days so that the second forecast represents a

revision rather than a forecast issued almost concurrently with the initial forecast. The average length of

time separating the two forecasts in our sample is about 47 days.

The purpose of the second criterion is to prevent “stale” forecasts (i.e., those that are not revised

following the previous quarter’s earnings announcement) from being included in the data. The third

criterion is an attempt to ensure that the latest forecast is not “contaminated” by knowledge of the actual

earnings number.

The total number of firm-quarters in the sample is 76,265 (containing at least twice as many

individual forecasts since our test design requires that there be both Fearliest and Flatest for each firm-

quarter), spanning the period from January 1983 to December 1997. The number of firm-quarters

increases steadily from an average of about 500 per fiscal quarter in the first five years of the sample

period to over 2,000 per fiscal quarter in the last five years of the period.

We repeat the tests, using consensus forecasts to define the expectation paths. In so doing, we

impose the restriction that each forecast reflect at least two individual analysts' forecasts in order to

ensure that there really is a “consensus” forecast rather than that of an individual analyst. The results

were essentially the same as those when the expectation paths were based on individual analysts'

forecasts. For the sake of parsimony, we present only the results based on the individual forecasts.

10

Actual earnings numbers were retrieved from the I/B/E/S database. Other financial accounting

data were retrieved from Compustat. In those instances where the I/B/E/S earnings number differed

substantially (by more than 50%) from the earnings number reported by Compustat and the difference

could not be explained by a special item (since I/B/E/S reports an “adjusted” earnings number), we

eliminated the observation. Return data were obtained from the Center for Research on Security Prices

(CRSP) database.

5. Tests and Results

5.1. Forecast revision paths and the earnings surprise

To ascertain whether our sample is comparable to those employed by previous research with

respect to the time series pattern of MBE, we produced the distribution of earnings surprises over time.

Our results (not presented here) show that both meeting and beating expectations have become more

prevalent in recent years, as documented by previous research (see, for example, Brown [1997]) as well

as current studies (see Brown [2000] and Lopez and Rees [2000]). Specifically, we find that the

proportion of unfavorable earnings surprises dropped from about 48% in the years 1983-1993 to only

31% in the more recent period of 1994 to 1997. Over these subperiods, the relative frequency of

meeting earnings expectations (i.e., a zero surprise) or beating them (a positive surprise) increased from

12% and 40% to 19% and 50%, respectively.

5.2. Tests of expectation management

The results of the tests of hypotheses 1a and 1b are provided in table 1. As shown in the table,

the percentage of negative earnings surprises over the entire sample is 43.08%, which is significantly

smaller (at the 1% confidence level, using the test of proportions) than the percentage of negative

forecast errors, 50.68%.

We further examine the change in the frequency of negative earnings surprises in two

subperiods, 1983-1993 and 1994-1997. These two subperiods are also used in other analyses in the

paper. Their selection is motivated by the introduction of First Call forecasts in the early 1990s and their

first appearance on the Web in 1994, developments that are likely to have widened the dissemination of

analyst forecasts and increased their use as a benchmark for firm performance.1 The conclusions and

1 Note also that the emergence of the “expectations game” seems to have taken place in the mid-1990s. A search of the key words “met expectations” or “beat expectations” in the financial turns up revealed a sharp increase around that time. Further, Brown [1997] and [2000], as well as our unreported results, show that the average analyst forecast error has become negative in recent years.

11

inferences are essentially the same when other partitions of the period (such as dividing it into two

equal-length subperiods) are used.

The difference between the relative frequencies of negative forecast errors and negative earnings

surprises increases from the early subperiod to the second subperiod, from 4.96% to 11.71%. This

increase in the difference is also significant at the 1% level, leading to the rejection of Hypothesis 1b.

These results are consistent with the expectation management hypothesis. This conclusion is

reinforced by the tests of Hypothesis 2a, whose results are provided in Table 2. The table shows the

proportion of firm-quarters with a negative forecast error that end with a positive surprise, and the

proportion of firm-quarters with a positive or zero error that end with a negative surprise. Observations

that belong to the first group are more likely to result from expectation manipulation than those in the

second group. Hypothesis 2A is based on the difference between these two proportions. As the table

shows, 35.68% of the firm-quarters with a negative forecast error ended, nonetheless (as a result of a

sufficiently large downward revision in earnings forecasts), with a positive earnings surprise. In

contrast, only 15.85% of the quarters with a positive, or zero, forecast error ended (due to a forecast

revision that “spoiled” what otherwise could have been a positive earnings surprise) with a negative

earnings surprise. This difference, which is statistically significant (at the 1% significance level, using

the test of proportions), suggests the presence of expectation management and the rejection of

Hypothesis 2a. The table also shows that the difference between the above relative frequencies, which is

19.83% for the entire period (35.68% - 15.85%), increased sharply from 10.41% (29.93% - 19.52%) in

the first subperiod to 32.16% (44.58% - 12.42%) in the second subperiod. These results suggest a

stronger propensity to manage expectations in recent years (a rejection of Hypothesis 2b).

All of the above findings are consistent with revisions in earnings forecasts being managed so as

to result in MBE at the end of the period. In particular, downward revisions are encouraged when, in

their absence, the earnings surprise is expected to be negative while upward revisions are discouraged if

they are expected to lead to a negative earnings surprise.

5.3. The reward to MBE

The evidence that expectations are managed implies that managers believe that there is a reward

to this activity in the form of a premium to MBE. To test for the existence of a premium to MBE

(hypothesis H3a), we measure the incremental quarterly abnormal return of instances in which

expectations are being met or beaten after controlling for the magnitude of the quarterly forecast error.

12

The return measure that we use is the beta-adjusted cumulative abnormal return, CAR, over the

period beginning one day prior to the date of Fearliest and ending one day after the release of the quarter’s

results.2

In testing H3a, we control for the magnitude of the forecast error by placing firm-quarters within

each error-sign group into portfolios based on the size of the forecast error. Using 5% intervals, this

results in nine equal-error-size portfolios for each of the positive and negative error groups, and one

portfolio for the zero error group.3

Hypothesis H3a is also tested (along with hypotheses H3b, H4, and H5,) by estimating the

following regressions:

CARi,Q = β0 + β1DMBEi,Q + β2DBEATi, Q +β3ERRORi,Q + β4SURPi,Q+β5DBEATi, Q*SURPi, Q+εi,Q, (1)

where i is the firm index and Q is a quarter notation. DMBE and DBEAT are dummy variables that

receive the value of 1 if, respectively, SURP ≥ 0 and SURP > 0. Otherwise, these variables receive the

value of zero. The overall forecast error for the quarter, ERROR, and the end-of-quarter earnings

surprise, SURP, are measured as described above and deflated by the firm's stock price at the beginning

of the quarter.

We expect β3 to be positive and significant, in line with the findings of the vast body of research

on the information content of earnings. Under the null of H3a, the coefficients β1, and β4 are not expected

to be significantly different from zero. Similarly, under the null of H3b the coefficients β1, and β4 are not

expected to significantly vary over time. Under the null of H4, β5 should not be different from zero and

under the null hypothesis H5, β2 should not be significantly different from zero.

Table 3 reports the results of testing Hypothesis 3a (premium to MBE) and Hypothesis 5

(differential premium to beating versus merely meeting expectations). The table presents the period

abnormal returns by path, controlling for the period’s forecast error. As noted earlier, this control is

obtained through the construction of equal error-size portfolios, in 5% increments.4 The table shows that

within almost every error-size portfolio, the period abnormal return, CARQ, associated with paths

2 We calculate several alternative measures of “abnormal return” for a period: the cumulative beta-adjusted abnormal return (which assumes daily re-balancing) over the period, the period’s “buy-and-hold” beta-adjusted abnormal return, and the period’s cumulative size-adjusted returns. All three measures led to essentially the same results. In addition, to account for return intervals of different length, we also used an average "per-day" measure of abnormal return. The use of this measure did not materially alter the results. 3 The portfolios are based on the percentage forecast errors, (EPS – Fearliest)/|EPS|, where EPS is the actual earnings per share. 4 Forming portfolios based on smaller error increments yields essentially similar results.

13

ending with a favorable earnings surprise (•-Up paths) is significantly higher than that associated with

paths ending with an unfavorable earnings surprise (•-Down paths).5

Panel B of table 3 summarizes the path results in Panel A across portfolios. Each value in this

summary panel is the simple average of the CARQ pertaining to the expectation path, across the nine

equal error-size portfolios.6 As the results for the positive error cases show, after controlling for the

magnitude of the forecast error for the quarter, the •-Up paths are associated with the highest CARQ,

having an average CARQ of 0.073 as compared with 0.040 for the Up-Zero path and 0.028 for the Up-

Down path. Similar results are obtained for cases with a negative forecast error, with the Down-Up path

having an average CARQ of -0.028 while the •-Down paths and the Down-Zero path have considerably

lower CARQ, of –0.047 and –0.074, respectively. Likewise, within the zero-error portfolio, the CARQ

for the expectation path that ends with a positive earnings surprise (Zero-Up) is larger by 1.3% (0.010

versus –0.003) than that ending with a negative earnings surprise7. Across all error groups, the average

CARQ for paths ending with a favorable surprise is greater by almost 3.0% than those ending with an

unfavorable surprise. The above differences in CARQ are significant at the 1% significance level or

higher, using the paired-difference t-test. Our findings (not shown here) also show the existence of

premium to MBE in each of the four fiscal quarters and for different firm-size portfolios. The variation

of the premium to MBE across quarters and firm size groups is not statistically significant.

Similar findings are reached when Hypotheses 3a and 5 are tested using regression (1). As the

results for the full sample indicate, the coefficients β1, and β4 are positive and significant, suggesting

that the earnings surprise affects the return for the quarter, even after controlling for the overall

quarterly forecast error (which is, as expected, a significant variable).8 The regression results also

suggest that beating expectations is associated with a higher return than just meeting expectations. The

coefficients of DBEAT (β2) and DBEAT*SURP (β5) are positive and significant.

5 There is a significant difference (at the 0.01 level or higher) between the average CARQ for the •- Down paths and that of the •- Up paths for 16 of the 18 portfolios in Panel A. 6 For the zero-error cases there is, obviously, only one error-size (EPS-Fearliest=0) portfolio. 7 Finding a premium to the Down-Zero path relative to the Up-Down or Zero-Down paths contradicts, ostensibly, the result of Soffer, Thiagarajan and Walther (STW) [2000]. They find that firms that preannounce negative news experience greater negative returns over the period leading to and including the earnings release date. However, the results of the two studies are not strictly comparable for at least two reasons. First, the number of observations in STW of negative preannouncements and no (neutral) earnings surprise is only about 200 (see their table 4) relative to almost 4,000 cases upon which our result is based. Also, unfavorable preannouncements are likely to represent cases where there is extremely bad news. Such extreme news may overshadow even subsequent positive earnings surprises. 8 The firm-quarter observations are not strictly independent because of the presence of multi-quarter observations for each firm. The regressions presented in this table were also estimated from a sub-sample in which a single quarter was

14

To test Hypothesis 3b concerning the equality of the premium over time, we divide the period

into the two subperiods (1983-1993 and 1994-1997) and test regression (1) separately for each of the

subperiods. The results reported in table 4 lead to the rejection of H3a. While the coefficient of the

intercept dummy (DMBE) in the early and recent subperiods is very similar (0.016 versus 0.017), the

slope coefficient for SURP is larger in the more recent years (0.675 versus 0.575 for the earlier years).

This difference, however, is not statistically significant (using the F test).

The null of H4, namely, that the premium arising from meeting or beating expectations is

identical to the penalty for failing to meet expectations is rejected by the data. As table 4 indicates, the

coefficient of the interactive variable DBEAT*SURP, β5, estimated for the full sample (and for each of

the subperiods), is positive and significant. This suggests that the reward (in terms of CAR) to a unit of

a favorable earnings surprise is greater than the penalty to an unfavorable earnings surprise.

Table 4 also shows that the premium to MBE is not confined to any fiscal quarter. The

coefficients of DMBE and SURP (β1 and β4) are positive and significant for each of the quarters, and not

significantly different across quarters.

The same results concerning the premium to MBE are obtained when the variables ERROR and

SURP are measured based on consensus analysts’ forecasts instead of individual analysts’ forecasts. For

the sake of brevity, the results are not presented here.

5.4. The reward to MBE as a function of earnings persistence and MBE persistence

Hypotheses H6, H7 and H8 predict differential premiums to MBE for, respectively, loss versus

profit cases, firms reporting earnings decreases versus those reporting earnings increases, and cases of

“habitual beaters” versus “sporadic” beaters.” These are tested using the following regression (quarter

and firm notations are omitted):

CAR = δ0 + δ1DMBE + δ2DMBEsubset + δ3ERROR + δ4SURP + δ5DMBE*SURP +

δ6DMBEsubset*SURP + ε. (2)

CAR, ERROR and SURP are as defined in regression (1). DMBE is a dummy variable that

receives the value of 1 if SURP ≥ 0 and 0 otherwise. DMBEsubset is a dummy variable that receives

the value of 1 if SURP ≥ 0 (i.e., DMBE =1) and, in addition, the case belongs to the subset of

observations to which the specific hypothesis refers.9 Depending on the hypothesis being tested, we

randomly drawn from each of the approximately 6,000 distinct firms in the sample. The results, and in particular the significance of the variables of interest, are similar to those obtained for the full sample. 9 Note that while the dummy variable DMBE in regression (2) is different from the dummy variable BEAT in regression (1), the interactive variables DMBE*SURP in regression (2) and BEAT*SURP in regression (1) are identical. Both are equal to SURP when SURP >0 and to 0 otherwise.

15

define that subset alternately as consisting of firm-quarters with a loss (H6), firm-quarters with an

earnings decrease (H7), and firm-quarters belonging to “habitual” beaters.10 For the purpose of testing

H7, earnings decreases are defined relative to the same quarter last year.11 In testing H8, “habitual

beaters” are defined as those firms with at least 20 quarters of forecast data that meet or beat analysts’

earnings forecasts for at least 75% of the quarters.

A differential premium to MBE in case of a loss (or an earnings decrease or a habitual

beater) will be reflected in δ2 and δ6 that are significantly different from zero. The results from testing

hypothesis H6 and H7 are provided in Table 5. As line 1 indicates, H6 can be rejected: The coefficient δ2

of the intercept dummy, DMBEsubset (the subset consisting of loss cases), is positive and significant (and

δ6, the coefficient of the related interactive variable, is positive and close to being significant), indicating

that the premium to MBE is more pronounced when the announced earnings are a loss. While the

magnitude of the coefficients suggests that the excess premium is economically trivial, the finding of a

greater premium to MBE in loss cases is consistent with the notion that a smaller than anticipated loss

has a greater valuation implication than a larger than anticipated profit. This may be due to the fact that

MBE in these cases indicates a reduction in the likelihood of bankruptcy or a perceived turning point in

the earnings pattern.

The results from testing H7 are shown in line 2 of Table 5. The coefficients δ2 and δ6 that are

associated with DMBEsubset, where the subset is defined as cases with earnings decreases, are not

statistically different from zero, suggesting that the premium to MBE is not a function of whether the

quarterly earnings represent an increase or a decrease relative to the same quarter last year. We also

tested H7 using the previous quarter as the benchmark for identifying earnings increases or decreases,

reaching essentially the same conclusions.

As line 3 of Table 5 shows, both δ2 and δ6 are positive and significant, indicating that the

premium to MBE is significantly higher for habitual beaters, thus rejecting H8. That is, rather than

discounting the favorable earnings surprises of firms that consistently produce such surprises, investors

value them even more. This finding is consistent with the findings of Kasznik and McNichols [1999]

that the premium is more pronounced for “habitual beaters” (in fact, they do not find a significant

premium to firms that only sporadically experience MBE) as well as the “momentum” story and the

related findings by Barth et al. [1999].

5.5. The reward to MBE as a function of management intervention 10 To illustrate, in testing H6, DMBEsubset is set equal to 1 for all MBE cases associated with a loss, and zero otherwise.

16

In order to test hypotheses relating to management intervention (H9 and H10), we estimate

regression (2), setting DMBEsubset equal to 1 alternately for MBE observations that are more likely

to represent expectation management and for MBE observations that are more likely to be driven by

earnings manipulation.

In line with the analysis in section 3.1, MBE cases that are more likely than others to result

from expectation management are identified as cases with a negative forecast error that end with a

zero or positive surprise (e.g., negative forecast error cases with Down-Zero or Down-Up paths). If

investors detect expectation management and, further, do not assign a premium, or assign a lower

premium, to cases where the MBE is obtained through expectation management (i.e., a rejection of

H9), δ2 and δ6 in regression (2) are expected to be negative.

A similar approach is used to test the association between the premium to MBE and earnings

management. Specifically, this association is tested by estimating regression (2), setting DMBEsubset

equal to 1 for all cases in which the MBE is more likely to have been driven by earnings

management. Such MBE cases are identified through two alternative procedures, both of which are

based on the identification of expected and unexpected accruals. Unexpected accruals are assumed

to be discretionary.

The first procedure to derive the unexpected discretionary accruals is based on a model that

relates the accruals each period to the level of activity (measured by revenues) and investment in

property plant and equipment) proposed by Jones (see Jones [1991] and Dechow et al. [1995]). In

the second procedure, unexpected discretionary accruals are defined as total accruals less working

capital accruals and depreciation. Working capital accruals are defined as:

Working Capital Accruals = ∆ Accounts Receivable + ∆ Inventories + ∆ Prepaid Expenses - ∆ Accounts Payable - ∆ Taxes Payable

Extracting depreciation, amortization and the working capital accrual components from total

accruals results in accruals consisting primarily of such items as loss and bad debt provisions (or

their reversal), restructuring charges, the effect of changes in estimates, gains or losses on the sale of

assets, asset write-downs, the accrual and capitalization of expenses, and the deferral of revenues

and their subsequent recognition. We refer to these accruals as “discretionary accruals” since their

amount or timing is usually discretionary. In other words:

Discretionary Accruals = Total Accruals – Working Capital Accruals – Depn. and Amort. 11 Defining earnings decreases relative to the previous calendar quarter led to essentially the same results.

17

Discretionary accruals for each firm-quarter derived using the above model (hereafter, the

“extraction model”) are scaled by the firm’s total assets at the beginning of the quarter; their

expected level is estimated as the mean of this ratio over all available quarters. Unexpected

discretionary accruals are measured for each firm-quarter as the difference between the firm’s

discretionary accruals in the quarter and the product of the above mean ratio and total assets of the

firm at the end of that quarter.

To examine the effect that earnings management might have on the premium to MBE, we

adjust the reported earnings of all the MBE cases by subtracting from the reported earnings the

amount of unexpected discretionary accruals (measured by the two alternative models described

above). We then recompute the earnings surprise (SURP) for all MBE cases to establish whether or

not they still retain their MBE designation after adjusting for unexpected discretionary accruals. We

then test H10 by estimating regression (2) setting DMBEsubset to 1 if the above cases retain their MBE

designation without the “help” of unexpected discretionary accruals.

If investors detect earnings management and, further, do not assign a premium, or assign a

lower premium, to cases where the MBE is obtained through earnings management (i.e., a rejection

of H10), then δ2 and δ6 are expected to be negative.

The results of the test of the association between expectation management and the premium

(H9) are provided in line 4 of Table 5. The results are based on estimating regression (2) with

DMBEsubset set equal to one for all MBE cases where MBE is likely to have been obtained through

expectation management and to zero otherwise. As the table shows, the coefficients δ2 and δ6 are

negative with δ2 being significantly negative. This result suggests that the premium to MBE is

significantly lower in instances in which the MBE is more likely to have been driven by expectation

management. Overall, however, the premium to MBE in these cases still exists and is lower by only

a small amount from the premium to MBE in other cases. This finding may either reflect investors’

inability to discern expectation management or their perception that MBE is a signal about future

performance of the firm, independently of how it was accomplished. At any rate, we hesitate to

draw strong conclusions from the small effect of expectation management on the premium since it

may reflect a weakness of our design in identifying instances where expectation management takes

place. Lines 5 of Table 5 show the results from estimating regression (2) with DMBEsubset set equal

to one for all MBE cases in which MBE is likely to have been obtained through earnings

management and to zero otherwise. Since earnings management can be detected through accruals,

18

an assessment regarding the presence or absence of earnings management can be made only when

the full quarterly report is publicly available. For this reason, the CAR accumulation period used to

test the earnings management-related hypothesis (H10) is extended to 5 days after the latest

allowable filing date of the 10Q report, or 50 days after the end of the quarter. Because the fourth

quarter’s results are published relatively late as part of the annual 10-K (up to 90 days after the end

the fiscal year), we elected to drop the fourth quarter observations from the analysis rather than

extend the accumulation period further.

Regression (2) is estimated twice, under the two alternative ways of estimating unexpected

accruals (the Jones model and the "extraction model") and thus of identifying MBE cases where

earnings management is less likely to have occurred. Both models lead to the same conclusion: The

premium to MBE in cases where the MBE is achieved only due to sufficiently positive unexpected

accruals is significantly lower than the premium to MBE in other cases.

The above finding suggests that investors are capable of discerning the effect of earnings

management on the earnings surprise and somewhat discount the resulting surprise. Yet, the extent of

the discount is economically minor (the coefficients δ2 and δ6 are very close to zero). Like the test on the

effect of expectation management on the premium to MBE, the small discount to the premium in cases

of earnings management could be due to the difference in the power of the methodology we use to

detect expectation management. In addition, while the CAR accumulation period used in this analysis

extends to 5 days after the filing date of the 10Q report, investors may not complete their assessment

regarding the presence of earnings management within that short period.

5.6. The information content of forecast revisions

The finding of a market reward to the expectation path is consistent with the notion that

investors assign less weight to analysts’ forecast revisions made during the quarter than to earnings

surprises at the earnings announcement time. We test this implication by decomposing the forecast

errors into its two components – the earnings forecast revision and the earnings surprise. We assess the

incremental contribution of each to the period abnormal return, by estimating the following regression

(quarter and firm notations are omitted):

CAR = α + β1REV + β2SURP +ε, (3)

where CAR and SURP are as defined in regressions (1). The revision, REV, defined in section 3, is the

overall forecast revision during the quarter measured as the difference between the latest forecast in the

quarter (Flatest) and the initial forecast (Fearliest). The results for all paths, shown in the first line of table 6,

suggest that while revisions in analysts’ forecasts are a significant factor in explaining the period return,

19

the effect of the surprise is significantly greater than that of the revision (at the 1% significance level).

While the coefficient of the surprise (β2) is 1.54, the coefficient of the revision (β1) is only 0.73.

Note that the two independent variables in regression (3) may be positively correlated, making

the detection of the contribution of each to the period’s return difficult. To alleviate this problem, we

separate the observations into two groups: those cases where the revision and surprise form a

“monotonic” path (i.e., the revision and surprise are in the same direction) and those cases where they

form a “nonmonotonic” path. We then re-estimate regression (3) for each group separately. The results

for quarters with a nonmonotonic path are even stronger than the results for all cases. The earnings

surprise associated with such paths clearly dominates the quarter’s abnormal return, as evidenced by the

relative magnitude of the coefficients on SURP (the earnings surprise) and REV (the forecast revision);

the coefficient on SURP is almost three times larger than the coefficient on REV (2.12 vs. 0.79).

The finding of a greater weight assigned by investors to earnings surprises than to forecast

revisions is reinforced when we compare the stock price reaction to forecast revisions with the response

to earnings surprises. The results reported in table 7 show that, after controlling for the magnitude of the

revision and the surprise, the stock price response to earnings announcements is, on average, 1.5 times

stronger than the response to analysts’ forecast revisions. We further find (not reported here) that

analyst’s revisions are less likely to trigger a revision in next year’s earnings forecasts than are earnings

surprises. The revision in next year’s earnings triggered by an earnings surprise is 1.8 times greater on

average than the revision triggered by a forecast revision for the current quarter, controlling for the

magnitude of the revision and the surprise. Both of these findings are consistent with the results from

regression (3) in which the coefficient for the earnings surprise is twice as large as the coefficient for the

revision (1.54 vs. 0.73, see table 6).

The lower weight assigned by investors to revisions in earnings estimates relative to earnings

surprises could rationalize an MBE strategy by the firm, whereby negative earnings forecast revisions

are induced in order to obtain subsequent favorable earnings surprises.12

5.7. Measurement errors in analysts’ forecasts

12 A possible explanation for the heavier weight assigned by investors to the earnings surprise relative to the interim revisions in earnings forecasts is that forecast revisions are more affected than actual earnings by events that have a transitory effect on future earnings. For example, interim revisions in analysts’ forecasts might often be triggered by management disclosures relating to the sale of assets, restructuring, layoffs and other publicized transitory events. We tested this conjecture by examining the effect that a forecast revision and an earnings surprise of an equal size in year t have on analysts’ earnings forecast for year t+1. The results (not shown) are consistent with this conjecture and thus with the more pronounced market response to earnings surprises than to earnings revisions.

20

One possible explanation for the dominance of the earnings surprise in explaining the entire

period returns is that investors overreact to earnings surprises. Another explanation is that earnings

surprises are informative in that they provide an indication of future firm performance. We test these

two explanations below. However, before testing these explanations, we attempt to rule out the

possibility that the results are driven by errors in estimating the unobservable market expectations of

earnings. In general, I/B/E/S forecasts are a good surrogate for the unobservable market expectations of

earnings (see Fried and Givoly [1982]; Brown et al. [1987]). Still, certain measurement errors in

analysts' forecasts could be correlated with the observed premium to MBE and thus reduce the precision

of identifying the sign (and magnitude) of the forecast error and the earnings surprise. We tried to assess

the impact of potential measurement errors in two ways. First, we replicate the tests substituting

mechanical forecasts (specifically, an AR(1) model in seasonal differences (see Foster [1977])13 for the

beginning-of-the-quarter I/B/E/S forecasts. In the second test we eliminate cases where the magnitude

of the revision is minor relative to the magnitude of the error, leading potentially to an incorrect

identification of the expectation path. The results from these two tests (not reported here) lead

essentially to the original findings, suggesting that these results are unlikely to be driven by

measurement errors.

5.8. The overreaction explanation

The incremental abnormal return for meeting or beating analysts’ expectations may also be yet

another manifestation of investors’ overreaction, a phenomenon that has been documented by past

research (see, for example, De Bondt and Thaler [1987] and Seyhun [1990]). For this explanation to

hold, some reversal of the announcement period abnormal return must occur in subsequent periods.

We examine the abnormal return across the equal-error portfolios of each expectation path over

the following quarter and for longer periods of one, two and three years subsequent to the earnings

announcement period. If reversal occurs, we would expect to find that the paths ending with a positive

earnings surprise would show a lower return in these subsequent periods. As is evident from the results

reported in table 8, there is no apparent reversal of the premium to MBE cases in the periods that

follows the earnings announcement. The results thus do not support the notion of investors’

overreaction to earnings surprises.14

13 We estimate the following regression from the most recent 10 years: (EPSQt - EPSQt-4) = δ + φ(EPSQt-1 - EPSQt-5) + εt, leading to the prediction EPSQt = EPSQt-4 + δ + φ(EPSQt-1 - EPSQt-5). 14 As indicated in section 3.2, we measure “abnormal return” for a period as the cumulative beta-adjusted abnormal return over the quarter. Previous research suggests that this measure provides biased values when accumulated over longer periods, such as a year (see, for example, Barber and Lyon [1996, 1997] and Kothari and Warner [1997]). The abnormal returns resented in table 8 are corrected for this bias using the procedure suggested by Barber and Lyon [1996, 1997].

21

5.9. The information content explanation

An explanation for a premium to MBE that is consistent with investors' rationality is that the

firm’s success in meeting or beating its earnings estimates (i.e., the expectation path) is informative

with respect to the future performance of the company. To test this explanation, we examine the

association between the incidence of MBE and the firms’ performance in subsequent quarters and years.

Firm performance is gauged by several accounting performance measures: return-on-assets, return-on-

equity, prevalence of losses, and sales and earnings growth. The results, shown in table 9, are consistent

with the MBE having information content with respect to future performance. Specifically, after

controlling for the forecast error, firms that meet or beat analysts’ earnings forecasts in a given quarter

exhibit significantly better performance over the following two years than firms that fail to meet

earnings expectations.

Table 9 shows that for cases with positive quarterly earnings forecast in year t, the return-on-

assets in the following year (ROAt+1), the return on equity (ROEt+1) and the growth rate of sales and net

income of firms that beat expectations is 5.7%, 11.4%, 23.0% and 49.2% respectively. In contrast, the

values for these performance measures are only 2.9%, 1.5%, 16.5% and 7.1%, respectively, for firms

with the same forecast error that failed to do so. Other performance measures in year t+1, such as the

operating margin and the percentage of losses, show the same superior performance of firms whose

quarterly earnings in year t beat analysts’ forecasts as compared with those who fell short of these

forecasts. Cases with negative quarterly earnings forecasts show the same pattern. Finally, as the table

shows, the better performance of the “expectation beaters” extends also to year t+2. All of the above

differences in performance between firms that meet or beat earnings expectations and those that do not

are significant at the 5% significance level or better.

These results are consistent with the better performance of habitual beaters of earnings

expectation documented by Kasznik and McNichols [1999]. To the extent that “buy” recommendations

are associated with anticipated positive firm performance, our results are also in line with the finding of

a positive correlation between MBE and “buy” recommendations reported by Abarbanell and Lehavy

[2000].

6. Conclusions The paper examines the recent phenomenon of the “expectation game” whereby companies and

investors focus on the degree to which reported earnings meet or beat analysts’ estimates. Anecdotal

and empirical evidence, including evidence provided by this paper, suggest that firms have become

more successful in MBE and that this success is achieved in part by managing expectations. The

22

evidence further shows that, after controlling for the absolute earnings performance, firms that manage

to meet or beat their earnings expectations, even at the expense of an earlier dampening of those

expectations, enjoy a higher return than their peers that fail to do so. While investors appear to apply

some discount to MBE cases that are likely to result from expectation or earnings management, the

premium in these cases is still significant. This finding, while rationalizing an MBE behavior by firms,

raises the question of investor rationality.

Further analyses indicate, however, that the emphasis placed by investors on the earnings

surprise is justified on economic grounds. The earnings surprise is more informative than the revisions

in earnings forecasts in predicting future earnings. Specifically, we find that earnings surprises are more

likely to trigger revisions in future annual earnings than are quarterly earnings revisions. Moreover,

earnings surprises are a reliable predictor of the firm’s future performance, after controlling for the

earnings performance. Firms whose quarterly earnings releases constitute a favorable surprise show, in

subsequent years, a higher growth in sales and earnings and a higher ROA and ROE than firms with the

same earnings performance but with unfavorable earnings surprises.

One of the interesting questions that still remains unanswered by the findings is why analysts do

not correct their forecasts for what appears to be a systematic downward bias in their late-in-the-period

forecasts. Or, to put it in more concrete terms, how could analysts underestimate Microsoft's quarterly

earnings 47 times in a row?

23

References Abarbanell J. and R. Lehavy, 2000, "Can stock recommendations predict earnings management and analysts' earnings forecast errors?" Working Paper, University of Northern Carolina. Barber, B. M. and J.D. Lyon, 1996, Detecting abnormal operating performance: The empirical power and specification of test statistics, Journal of Financial Economics 41, (3): 359-399. ___________ and _______, 1997, Detecting long-run abnormal stock returns: The empirical power and specification of test statistics, Journal of Financial Economics 43, (3): 341-372. Barefield, R. M. and E. E. Comiskey, 1975, The accuracy of analysts’ forecasts of earnings per share, Journal of Business Research, 3 (3): 241-252. Barth, M., Elliott, J.A. and M.W. Finn, 1999, Market rewards associated with patterns of increasing earnings, Journal of Accounting Research, 37 (2): 387-413. Brown, L. D., 1997, Analyst forecasting errors: Additional evidence, Financial Analysts Journal, 53 (6): 81-88. __________ , 2000, Managerial behavior and the bias in analysts’ earnings forecasts, Working Paper, Georgia State University, Atlanta, GA. ___________, Griffin, P.A., Hagerman, R.L. and M.E. Zmijewski, 1987, An evaluation of alternative proxies for the market's assessment of unexpected earnings, Journal of Accounting and Economics 9 (2): 159-193. Brooks, L.D. and D.A. Buckmaster, 1976, Further evidence of the time series properties of accounting income, The Journal of Finance 31: 1359-1373. Burgstahler, D., and M. Eames, 1998, Management of earnings and analysts’ forecasts, Working Paper, University of Washington, Seattle, WA. DeBondt, W. F. M. and R.H. Thaler, 1990, Do security analysts overreact? American Economic Review, 80 (2): 52-57. Dechow, P.M., R.G. Sloan and A.P. Sweeney, 1995, Detecting Earnings Management, The Accounting Review, 70 (2): 193-225. Degeorge, F., J. Patel and R. Zeckhauser, 1999, Earnings management to exceed thresholds, Journal of Business, 72 (1): 1-33. Elgers, P.T. and M.H. Lo, 1994, Reduction in analysts’ annual earnings forecast errors using information in prior earnings and security returns, Journal of Accounting Research 32: 290-303 Foster, G., 1977, Quarterly accounting data: Time-series properties and predictive-ability results, The Accounting Review 52 (1): 1-21. Fox, J., 1997, Learn to play the earnings game (and Wall Street will love you), Fortune, 135, (6): 76-80. Francis, J., D. Philbrick and K. Schipper, 1994, Shareholder litigation and corporate disclosures, Journal of Accounting Research, 32 (2): 137-164.

24

Fried, D. and D. Givoly, 1982, Financial analysts' forecasts of earnings: A better surrogate for market expectations, Journal of Accounting and Economics 4, (2): 85-107. Hayn, C., 1995, The Information Content of Losses, Journal of Accounting and Economics, 20: 125-153. Jones, J.J., 1991, Earnings Management During Import Relief Investigations, Journal of Accounting Research, 29 (2): 193-228. Kasznik, R., 1999, On the association between voluntary disclosure and earnings management, Journal of Accounting Research: 37 (1): 57-81. ________ and B. Lev, 1995, To warn or not warn: Management disclosures in the face of an earnings surprise. The Accounting Review 70 (1): 113-134 ________ and M. McNichols, 1999, Does meeting expectations matter: Evidence from analyst revisions and share prices, Working Paper, Stanford University, Stanford, CA. Kothari, S. P. and J.B. Warner, 1997, Measuring long-horizon security price performance, Journal of Financial Economics 43 (3): 301-339. Lopez, T. and L. Rees, 2000, The effect of meeting analysts’ forecasts and systematic positive forecast errors on the information content of unexpected earnings, Working Paper, Texas A& M University, College Station, TX. Lyon, J. D., Barber, B.M. and Tsai, C, 1999, Improved methods for tests of long-run abnormal stock returns, Journal of Finance 54, (1): 165-201. McGee, S. 1997, As stock market surges ahead, “predictable” profits are driving it, The Wall Street Journal, C1, 5/5/97. Payne J. L. and S.W. Robb, 1997, Earnings management: The effect of ex ante earnings expectations. Working Paper, University of Mississippi. Seyhun, H. N., 1990, Overreaction or fundamentals: Some lessons from insiders' response to the market crash of 1987, Journal of Finance 45 (5): 1363-1388. Soffer, L.C., S. R. Thiagarajan and B. R. Walther, 2000, Earnings Preannouncement Strategies, Review of Accounting Studies, forthcoming. Skinner, D.J., 1997, Earnings disclosures and stockholder lawsuits, Journal of Accounting and Economics 23 (3): 249-282. Vickers, M., 1999, Ho-hum, another earnings surprise, Business Week, (May 24): 83-84. Zarowin, P., 1989, Does the stock market overreact to corporate earnings information? , Journal of Finance, (44): 1385-1400

25

EPSQ-1 Fearliest Flatest EPSQ

Beginning of End of Quarter, Q Quarter, Q

time

Legend: EPSQ-1 and EPSQ: the actual earnings announcements for quarters Q-1 and Q, respectively. Fearliest: the first forecast for quarter Q following the earnings announcement for quarter Q-1. Flatest: the last forecast for quarter Q prior to the release of the earnings announcement for quarter Q. ERRORQ: the overall forecast error for the quarter, measured as EPSQ – Fearliest. REVQ: the forecast revision for the quarter, measured as Flatest – Fearliest. SURPQ: the earnings surprise for quarter Q, measured as EPSQ – Flatest.

REVQ SURPQ

ERRORQ

Figure 1

26

Table 1 Relative Frequency of Negative Forecast Errors and Negative Earnings Surprises1

(1)

Percentage of negative earnings

surprises

(2) Percentage of negative

forecast errors

(3) = (2) – (1) Excess of negative earnings

errors over negative surprise cases

All Years 43.08% 50.68% 7.60% By Subperiod: 1983-1993 48.59% 53.55% 4.96% 1994-1997 30.69% 43.40% 11.71%

1 Earnings surprise: Difference between the actual earnings and the latest forecast for the quarter, EPS - Flatest. Forecast error: Difference between the actual earnings and the earliest forecast for the quarter, EPS - Fearliest.

27

T

able 2 Frequency of Selected E

xpectation Paths, by Period1

C

ases with a N

egative Forecast Error that End with:

(a)

Positive Surprise (D

own-U

p Path)

(b) Zero Surprise

(Dow

n-Zero Path)

(c) G

roup I (a) and (b) com

bined. I.e., C

ases with either

Positive or Zero Surprise(D

own-U

p or D

own-Zero Paths)

(d) C

ases with a

Positive Forecast Error that End w

ith a Negative

Surprise (U

p-Dow

n Path)

(e) C

ases with a Zero

Forecast Error that End w

ith a N

egative Surprise (U

p-Dow

n Path)

(f) G

roup II (d) and (e) com

bined. I.e., cases w

ith either a Positive or Zero

Forecast Error with a

Negative Surprise

(Up-D

own Path)

Difference betw

een G

roup I and Group

II in the proportion of cases w

here the sign of the Surprise is opposite the sign

of the Error

Period N

%

of all cases w

ith a negative

error

N

% of all

cases with a

negative error

N

% of all cases

with a

negative error

N

% of all

cases with

a positive error

N

% of all

cases with

a zero error

N

% of all

cases with

either a positive or zero error

[1]

[2] [3]

[4] [5]

[6] [7]

[8] [9]

[10] [11]

[12] [13] = [6] – [12]

All

Years

9,080

24.65%

4,061

11.02%

13,141

35.68%

4,715

14.19%

1,536

24.76%

6,251

15.85%

19.83%

*

1983-1993 5,023

22.45%

1,672 7.47%

6,695

29.93%

3,04717.69%

759

29.86%

3,80619.52%

10.41%

*

1994-1997

4,057 28.10%

2,389

16.52%

6,44644.58%

1,668

10.42%

77721.22%

2,445

12.42%

32.16%*

Difference

between

subperiods

14.65%

*

-7.10%

*

21.75%

*

1 Expectation paths are defined by the sign of the forecast revision and the earnings surprise. Forecast revision: D

ifference between the latest forecast and the earliest forecast for the quarter, F

latest – Fearliest .

Earnings surprise: Difference betw

een the actual earnings and the latest forecast for the quarter, EPS - Flatest .

Forecast error: Difference betw

een the actual earnings and the earliest forecast for the quarter, EPS - Fearliest .

* Significant at the 1%

level, using the test of proportions.

28

Table 3 Mean Quarterly Abnormal Returns (CARQ) by

Sign, Size of Forecast Error and Expectation Path1

PANEL A: By Sign and Size of Forecast Error, and Expectation Path

Positive Error Cases: EPS - Fearliest > 0 Negative Error Cases: EPS - Fearliest < 0 Size of Forecast

Error Path Based on

Direction of Revision and Surprise

No. of Obs.

Period Return, CARQ

Size of Forecast Error

Path Based on Direction of Revision and Surprise

No. of Obs.

Period Return, CARQ

Up-Down 1574 0.018 Up-Down 1238 -0.002 Up-Zero 995 0.009 Zero-Down 1006 -0.004

5.0%>X>0.0% Up-Up 614 0.027 0.0%>X>-5.0% Down-Zero 957 -0.008 Zero-Up 1966 0.032 Down-Down 347 -0.002 Down-Up 2516 0.034 Down-Up 2278 0.002

Total No. ; Weighted Return 7665 0.026 Total No. ; Weighted Return 5826 -0.002 Up-Down 1064 0.020 Up-Down 855 -0.026 Up-Zero 726 0.033 Zero-Down 671 -0.022

10.0%>X>5.0% Up-Up 1886 0.045 -5.0%>X>-10.0% Down-Zero 794 -0.027 Zero-Up 1537 0.047 Down-Down 916 -0.025 Down-Up 1930 0.041 Down-Up 1597 -0.011

Total No. ; Weighted Return 7143 0.039 Total No. ; Weighted Return 4833 -0.020 Up-Down 539 0.028 Up-Down 537 -0.034 Up-Zero 339 0.059 Zero-Down 328 -0.033

15.0%>X>10.0% Up-Up 1798 0.058 -10.0%>X>-15.0% Down-Zero 443 -0.040 Zero-Up 822 0.081 Down-Down 926 -0.037 Down-Up 1103 0.066 Down-Up 892 -0.015

Total No. ; Weighted Return 4601 0.061 Total No. ; Weighted Return 3126 -0.030 Up-Down 319 0.031 Up-Down 425 -0.048 Up-Zero 164 0.084 Zero-Down 238 -0.061

20.0%>X>15.0% Up-Up 1493 0.073 -15.0%>X>-20.0% Down-Zero 297 -0.065 Zero-Up 406 0.086 Down-Down 815 -0.046 Down-Up 642 0.062 Down-Up 660 -0.017

Total No. ; Weighted Return 3024 0.069 Total No. ; Weighted Return 2435 -0.042 Up-Down 215 0.031 Up-Down 293 -0.047 Up-Zero 89 0.036 Zero-Down 139 -0.076

25.0%>X>20.0% Up-Up 1030 0.093 -20.0%>X>-25.0% Down-Zero 150 -0.073 Zero-Up 258 0.086 Down-Down 687 -0.047 Down-Up 449 0.083 Down-Up 392 -0.027

Total No. ; Weighted Return 2041 0.081 Total No. ; Weighted Return 1661 -0.047 Up-Down 163 0.049 Up-Down 261 -0.043 Up-Zero 79 0.028 Zero-Down 112 -0.082

30.0%>X>25.0% Up-Up 885 0.112 -25.0%>X>-30.0% Down-Zero 143 -0.077 Zero-Up 185 0.101 Down-Down 636 -0.064 Down-Up 329 0.062 Down-Up 331 -0.036

Total No. ; Weighted Return 1641 0.090 Total No. ; Weighted Return 1483 -0.057 Up-Down 134 0.015 Up-Down 198 -0.042 Up-Zero 51 0.053 Zero-Down 105 -0.042

35.0%>X>30.0% Up-Up 624 0.112 -30.0%>X>-35.0% Down-Zero 114 -0.089 Zero-Up 140 0.079 Down-Down 519 -0.071 Down-Up 258 0.070 Down-Up 280 -0.042

Total No. ; Weighted Return 1207 0.086 Total No. ; Weighted Return 1216 -0.059

29

Table 3 (Continued) Mean Quarterly Abnormal Returns (CARQ) by

Sign, Size of Forecast Error and Expectation Path1

PANEL A (Continued) Positive Error Cases: EPS - Fearliest > 0 Negative Error Cases: EPS - Fearliest < 0

Size of Forecast Error

Path Based on Direction of Revision

and Surprise

No. of Obs.

Period Return, CARQ

Size of Forecast Error

Path Based on Direction of Revision and Surprise

No. of Obs.

Period Return, CARQ

Up-Down 73 0.014 Up-Down 161 -0.039 Up-Zero 23 0.007 Zero-Down 52 -0.054

40.0%>X>35.0% Up-Up 475 0.110 -35.0%>X>-40.0% Down-Zero 85 -0.147 Zero-Up 81 0.074 Down-Down 493 -0.065 Down-Up 160 0.093 Down-Up 175 -0.044

Total No. ; Weighted Return 812 0.092 Total No. ; Weighted Return 966 -0.063 Up-Down 556 0.046 Up-Down 2209 -0.068 Up-Zero 186 0.048 Zero-Down 861 -0.073

X>40.0% Up-Up 2341 0.108 X<-40.0% Down-Zero 950 -0.143 Zero-Up 376 0.119 Down-Down 7796 -0.114 Down-Up 828 0.081 Down-Up 1975 -0.066

Total No. ; Weighted Return 4287 0.093 Total No. ; Weighted Return 13791 -0.099

PANEL B: Summary by Sign of Forecast Error and Expectation Path Path Based on

Direction of Revision and Surprise*

No. of Portfolios (No. of Obs. in

Portfolios)

Mean CARQ Across Error-

size Portfolio

Difference Between CARQ for the •-Up and •-Down Paths

(t-statistic) Positive Error Cases: EPS – Fearliest>0 Up-Down 9 ( 4,637) 0.028 (n=32,492 firm-quarters) Up-Zero 9 ( 2,652) 0.040 0.045 • - Up 27 (25,203) 0.073 (3.85) Zero Error Cases: EPS - Fearliest = 0 Up-Down 1 (1,503) -0.003 0.013 (n=6,051 firm-quarters) Zero-Zero 1 (2,081) 0.008 (2.02) Down-Up 1 (2,467) 0.010 Negative Error Cases: EPS – Fearliest <0 • - Down 27 (22,824) -0.047 0.019 (n=35,337 firm-quarters) Down-Zero 9 ( 3,933) -0.074 (2.84) Down-Up 9 (8,580) -0.028 * (• is Up, Zero or Down)

1 Expectation paths are defined by the sign of the forecast revision and the earnings surprise. Forecast revision: Difference between the latest forecast and the earliest forecast for the quarter, Flatest – Fearliest. Earnings surprise: Difference between the actual earnings and the latest forecast for the quarter, EPS - Flatest. Forecast error: Difference between the actual earnings and the earliest forecast for the quarter, EPS - Fearliest. CARQ : Cumulative abnormal return over the quarter beginning on the day prior to the earliest forecast and ending the day following the earnings release.

30

Table 4 Results for Regression (1):

CARi,Q = ββββ0 + ββββ1DMBEi,Q + ββββ2DBEATi, Q +ββββ3ERRORi,Q + ββββ4SURPi,Q +ββββ5DBEATi,Q*SURPi,Q +

εεεεi,Q,

(t-statistics are provided in parentheses)

Sample β0 β1 β2 β3 β4 β5 R2 (%)

Full Sample (n=75,910) -0.028

(-25.48) 0.016 (7.60)

0.037 (14.05)

0.519 (9.83)

0.634 (8.18)

0.366 (5.14)

3.9

By Period Subperiod 1: 1983-1993 (n=42,371)

-0.028 (-17.87)

0.016 (5.41)

0.041 (12.46)

0.498 (7.70)

0.575 (5.39)

0.436 (4.19)

4.0

Subperiod 2: 1994-1997 (n=33,538)

-0.029 (-17.49)

0.017 (6.33)

0.052 (14.16)

0.588 (8.06)

0.675 (7.40)

0.732 (3.40)

3.9

By Quarter1 Quarter 1 (n=14,216)

-0.034 (-12.24)

0.023 (4.15)

0.040 (7.44)

0.651 (5.25)

0.306 (3.69)

0.397 (1.84)

4.5

Quarter 2 (n=15,233) -0.028 (-11.45)

0.019 (3.92)

0.035 (7.68)

0.574 (4.90)

0.631 (3.77)

0.476 (3.01)

4.1

Quarter 3 (n=16,444) -0.028 (-11.88)

0.018 (4.01)

0.034 (7.70)

0.608 (5.08)

0.662 (3.83)

0.295 (1.81)

4.0

Quarter 4 (n=17,521) -0.029 (-12.90)

0.019 (4.39)

0.035 (8.38)

0.421 (4.80)

0.518 (3.71)

0.355 (2.25)

4.6

Legend: CARQ : Cumulative abnormal return over the quarter beginning the day prior to the earliest forecast and ending the day following the earnings release.

ERROR: Forecast error computed as the difference between the actual earnings and the earliest forecast for the quarter, EPS - Fearliest, standardized by price at the beginning of the quarter. SURP: Earnings surprise computed as the difference between the actual earnings and the latest forecast for the quarter, EPS - Flatest, standardized by price at the beginning of the quarter.

DMBE (DBEAT): Dummy variable that takes on the value of 1 if SURP ≥ 0 and 0 otherwise

1 Only calendar-year firms were considered in this analysis.

31

Table 5 Results for Regression (2):

CARi,Q = δδδδ0 + δδδδ1DMBE + δδδδ2DMBEsubset + δδδδ3ERROR + δδδδ4SURP + δδδδ5DMBE*SURP+ δδδδ6DMBEsubset*SURP +εεεεi,Q

(t-statistics are provided in parentheses)

Line Tested Hypothesis (subset for which DMBEsubset =1 )

δ0 δ1 δ2 δ3 δ4 δ5 δ6 R2 (%)

1

H6 (Losses)1

-0.028

(-24.45)

0.045

(15.88)

0.004 (2.12)

0.530 (9.90)

0.695 (7.81)

0.096 (1.78)

0.249 (1.93)

3.5

2

H7 (Earnings Decreases)2

-0.028

(-24.55)

0.044

(16.45)

0.040 (1.76)

0.529 (9.88)

0.655 (7.53)

0.100 (1.81)

1.318 (0.61)

3.6

3

H8 (Habitual Beaters)3

-0.027

(-24.60)

0.044

(11.37)

0.034 (1.97)

0.525 (9.77)

0.663 (7.60)

0.235 (1.25)

1.216 (3.57)

3.5

4

H9 (Cases likely to represent expectation management)

-0.028

(-24.53)

0.044 (9.97)

-0.005 (-1.99)

0.529 (9.90)

0.655 (7.54)

0.096 (1.72)

-0.002 (-0.69)

3.5

5

H10 (Cases likely to represent earnings management – based on the Jones’ model)

-0.027

(-19.17)

0.046 (5.04)

-0.003 (-1.84)

0.568

(13.62)

0.614 (8.76)

0.342 (4.26)

-0.005 (-2.42)

3.7

6

H10 (Cases likely to represent earnings management – based on the “extraction model”)

-0.029

(-24.66)

0.044 (4.58)

-0.004 (-2.13)

0.601

(12.48)

0.579 (9.95)

0.282 (3.67)

-0.004 (-2.01)

3.8

Legend: CAR : In the regressions presented in lines 1-4: cumulative abnormal return over the quarter beginning the day prior to the

earliest forecast and ending the day following the earnings release. In the regressions in lines 5 and 6: cumulative abnormal return over the period beginning the day prior to the first forecast and ending 50 days following the end of the quarter.

ERROR: Forecast error computed as the difference between the actual earnings and the earliest forecast for the quarter, EPS - Fearliest, standardized by price at the beginning of the quarter.

SURP: Earnings surprise computed as the difference between the actual earnings and the latest forecast for the quarter, EPS - Flatest, standardized by price at the beginning of the quarter. DMBE: Dummy variable that takes on the value of 1 if SURP ≥ 0 and 0 otherwise. DMBEsubset: Dummy variable that takes on the value 1 if SURP ≥ 0 and, in addition, the case belongs to a designated subset of the sample. Otherwise DMBEsubset = 0.

1 Results are presented based on whether net income is positive or negative. Comparable results are obtained when income from continuing operations is used to classify the firm's profitability. 2 Earnings decreases are defined relative to the same fiscal quarter last year. 3 Habitual beaters are firms that beat or meet expectations in at least 75% of the quarters. Firms had to have at least 20 quarters of data to participate in this analysis.

32

Table 6

Results for Regression (3)1,2

CARI,Q = αααα + ββββ1REVi,Q + ββββ2SURPi,Q + εεεεi,Q (t-statistics are provided in parentheses)

Sample Intercept β1 β2 R2

Full Sample (n=76,265)

0.00(0.82)

0.73(11.98)

1.54 (32.18)

0.025

Observations with Monotonic Paths (Down-Down and Up-Up Paths)

0.00(-1.31)

1.07(8.13)

1.23 (12.11)

0.022

Observations with Non-Monotonic Paths (Down-Up and Up-Down Paths)

0.00(1.22)

0.79(7.01)

2.12 (19.24)

0.026

Legend: CAR : Cumulative abnormal return over the quarter beginning on the day prior to the earliest forecast and ending the day following the earnings release.

REV: Forecast revision computed as the difference between the latest forecast and the earliest forecast for the quarter, Flatest - Fearliest, standardized by price at the beginning of the quarter.

SURP: Earnings surprise computed as the difference between the actual earnings and the latest forecast for the quarter, EPS - Flatest, standardized by price at the beginning of the quarter.

33

Table 7 Cumulative Abnormal Returns Associated with Quarterly Forecast Revisions

and Earnings Surprises

Cumulative Abnormal Returns During: Magnitude of the Forecast

Revision or Earnings Surprise

(1) Forecast Revision Period1

(2) Earnings

Announcement Period1

(3) Ratio of (2) to (1)

Upward Revisions / Positive Earnings Surprises

5.0% > X ≥ 0.0% 0.004 0.008 2.00 10.0% > X ≥ 5.0% 0.011 0.016 1.45 15.0% > X ≥ 10.0% 0.016 0.026 1.63 20.0% > X ≥ 15.0% 0.019 0.029 1.53 25.0% > X ≥ 20.0% 0.021 0.036 1.71 30.0% > X ≥ 25.0% 0.024 0.033 1.38 35.0% > X ≥ 30.0% 0.030 0.042 1.40 40.0% > X ≥ 35.0% 0.034 0.051 1.50

X ≥ 40.0% 0.036 0.084 2.33 Downward Revisions/ Negative Earnings Surprises

0.0% > X ≥ -5.0% -0.007 -0.010 1,43 -5.0% > X ≥ -10.0% -0.012 -0.019 1.58

-10.0% > X ≥ -15.0% -0.017 -0.026 1.53 -15.0% > X ≥ -20.0% -0.023 -0.029 1.26 -20.0% > X ≥ -25.0% -0.028 -0.037 1.32 -25.0% > X ≥ -30.0% -0.029 -0.044 1.52 -30.0% > X ≥ -35.0% -0.034 -0.040 1.18 -35.0% > X ≥ -40.0% -0.037 -0.052 1.41

X ≤ 40.0% -0.040 -0.073 1.83 1 The earnings announcement (the revision) period is the 3-day period, including the day of the earnings announcement (revision) the preceding day and the following day.

34

Table 8 Cumulative Abnormal Returns Subsequent to the Quarterly Earnings Announcement

by Expectation Path

Mean Abnormal Return Across Error-size Portfolios No. of

Obs

In Quarter t

In Subsequent Periods1

Expectation Path . CARQ CARQ+1 CARyear t+1

CARyear t+2 CARyear t+3

Positive Errors: EPS - Fearliest > 0 Up-Down 4,637 0.025 0.007 0.012 0.007 0.010Up-Zero 2,652 0.032 0.003 -0.011 0.010 0.009Up-Up 11,146 0.079 0.017 0.018 0.015 0.020Zero-Up 5,771 0.059 0.012 0.015 0.020 0.016Down-Up 8,286 0.049 0.010 0.017 0.021 0.017 Zero Errors: EPS - Fearliest = 0 Up-Down 1,503 -0.003 0.008 -0.003 0.006 0.011Zero-Zero 2,081 0.008 0.005 0.015 0.003 --0.008Down-Up 2,467 0.010 0.009 0.016 0.015 0.014 Negative Errors: EPS - Fearliest < 0 Up-Down 6,177 -0.042 -0.009 0.013 -0.004 0.010Zero-Down 3,512 -0.038 -0.005 -0.009 0.011 0.013Down-Zero 3,933 -0.060 0.006 0.018 -0.010 -0.005Down-Down 13,135 -0.080 -0.020 -0.010 0.009 -0.007Down-Up 8,580 -0.035 0.004 0.017 0.013 0.012

Legend: CARQ : Cumulative abnormal return over the quarter beginning on the day prior to the earliest forecast and ending the day following the earnings release. (This column is from Panel B of Table 3.) CARQ+1: Cumulative abnormal return over the following quarter beginning two days after the earnings release date for quarter t and ending the day following the earnings release for quarter t+1. 1 The annual abnormal returns are adjusted for bias by the procedure suggested by Barber and Lyon [1996, 1997].

35

Table 9

Firm Perform

ance in Fiscal Years Subsequent to the Q

uarterly Earnings Surprise, by the Sign of the E

arnings Surprise1

ME

AN

VA

LU

ES

Y

ear t+1 Y

ear t+2 Path in

Fiscal Year t

RO

A

RO

E %

Losses

Sales G

rowth

M/B

Profit

Margin

Income

Grow

thR

OA

R

OE

%

Losses Sales

Grow

thM

/B

Profit M

arginIncom

e G

rowth

Positive Errors

Up-D

own

0.029 0.015

0.1630.165

2.4720.028

0.0710.033

0.0030.183

0.1232.542

0.030-0.041

Up-Zero

0.039 0.019

0.1410.209

2.7280.041

0.1130.046

0.0660.106

0.1862.856

0.0390.143

• - Up

0.057* 0.114*

0.102*0.230*

2.859*0.054*

0.492*0.057*

0.118*0.096*

0.207*2.857*

0.052*0.445*

Zero Errors

Up-D

own

0.044 0.078

0.1300.220

2.8950.041

0.0850.053

0.0860.106

0.1802.929

0.051-0.001

Zero-Zero 0.064

0.110 0.099

0.2493.390

0.0590.164

0.059 0.113

0.1290.222

3.4040.056

0.111 D

own-U

p 0.056** 0.109*

0.105**0.275*

3.178**0.053

0.323*0.058

0.123*0.124**

0.1962.975

0.0560.364*

Negative E

rrors

• - D

own

0.037

0.043 0.150

0.1722.573

0.0330.061

0.040 0.061

0.1550.122

2.4100.037

0.159 D

own-Zero

0.047 0.083

0.1230.225

2.6250.041

0.1840.044

0.0740.147

0.1602.648

0.0400.385

Dow

n-Up

0.041 0.124* 0.143

0.1902.555

0.0380.387*

0.047 0.125*

0.140 0.137**

2.667**0.044

0.426* • stands for U

p, Zero and Dow

n. * Indicates significant difference at the 0.05 level or better using a one-tailed t-test (see note 1) ** Indicates significant difference at the 0.10 level or better using a one-tailed t-test (see note 1) Legend: R

OA

: Return on assets m

easured as net income divided by total assets

RO

E: Return on equity m

easured as net income divided by the book value of equity.

% Losses: N

o. of firms reported a loss based on net incom

e. Sales G

rowth: G

rowth in sales revenues.

M/B

: Market value of equity divided by the book value of equity.

Profit Margin: N

et income divided by sales.

Income G

rowth: G

rowth in net incom

e. (Note: M

easures using net income w

ere repeated using Income from

Continuing O

perations with sim

ilar results.) 1For the Positive Error C

ases, differences between the perform

ance measure for the U

p-Dow

n and • - Up paths are com

pared. For the Zero Error C

ases, differences between the U

p - Dow

n and Dow

n-Up paths are com

pared. For the N

egative Error Cases, differences betw

een the • - Dow

n and Dow

n-Up paths are com

pared.